interesting discussion. Here at LABSS we are betting on the relevance
of "intelligent agents" from several years, even if the temptation to
reduce complexity sometimes emerges.
About the definitions - proactivity, sociality and - even more
controversial - autonomy, I have always taken them not as objective
requirements (actual autonomy requires *rolling drums* free will) but
as design guidelines. And they tend more to exclude solutions (spin
particles as agents) than to set precise requirements.
The paradigms (1 - collective decision making VS. 2 - simulation)
sound empty to me. Most of the simsoc work will fall under (2) and my
intuition says that they could me homeomorphic (by accurately defining
the decision to be taken, for example)
Peer-Olaf's agents seems quite "intelligent" to me. And I do not think
that definition struggles will help the field. However a comment on:
<<Ouragents perceive a goal in that they want to either buy something or return
something. For buying they have a sub goal; that they are trying to buy the
right thing. If they are not sure they will ask for help. Our agents are not
only reactive but also flexible, i.e. they are capable to recover from a
failure of action. >>
Sounds nice, even if I wonder if the goals are explicitly represented
or not. If not, well, that amounts to playing with words - so they
have also a goal of not buying the wrong thing, not (not return and
not buy) something, etc...
Lastly, examples of "intelligent" agents we have implemented are the
ones in Repage (http://jasss.soc.surrey.ac.uk/9/2/3.html) for
Reputation, and the norm reasoners that are still under development in
the EMIL project (http://emil.istc.cnr.it,
http://www.istc.cnr.it/doc/146a_2138p_FS0705AndrighettoG.pdf for more
In both cases our group is following an unorthodox approach, in the
spirit of Scott Moss recommendation of micro plausibility. We do not
apply a standard modellng paradigm for intelligent agents (BDI), but
concentrate on describing, as faithfully as possible, the theory into
the software mechanism.
And, yes, the simulation results are cumbersome and difficult to
interpret. Does that mean we should relinquish this path? Or perhaps
that we need better data analysis procedures?
LABSS-ISTC-CNR - http://labss.istc.cnr.it
eRep project - http://megatron.iiia.csic.es/eRep
ICORE 09 - http://www.reputation09.net/
Via San Martino della Battaglia, 44
tel. +39 06 44595 336
Please avoid sending me Word or PowerPoint attachments.